Activation of MAP Kinase Signaling Through ERK5 But Not ERK1 Expression Is Associated with Lymph Node Metastases in Oral Squamous Cell Carcinoma (OSCC)

Abstract

In an attempt to further elucidate the pathomechanisms in oral squamous cell carcinoma (OSCC), gene expression profiling was performed using a whole-transcriptome chip that contains 35,035 gene-specific 70mere oligonucleotides (Human OligoSet 4.0; Operon, Cologne, Germany) to a set of 35 primary OSCCs. Altogether, 7390 genes were found differentially expressed between OSCC tumor samples and oral mucosa. To characterize the major biologic processes in this tumor collection, MAPPFinder, a component of GenMAPP version 2.1, was applied to this data set to generate a statistically ranked list of molecular signaling pathways. Among others, cancer-related pathways, such as mitogen-activated protein (MAP) kinase signaling (z score = 4.6, P < .001), transforming growth factor-beta signaling (z score = 3.0, P = .015), and signaling pathways involved in apoptosis (z score = 2.1, P = .037), were found deregulated in the OSCC collection analyzed. Focusing on the MAP kinase signaling pathway, subsequent tissue microarray analyses by immunohistochemistry revealed an increase in protein expression of MAP kinase-related proteins ERK1 in 22.8% (48 of 209) and ERK5 in 27.4% (76 of 277), respectively. An association of high ERK5 but not of high ERK1 expression with advanced tumor stage and the presence of lymph node metastases was found (P = .008 and P = .016, respectively). Our analysis demonstrates the reliability of the combined approach of gene expression profiling, signaling pathway analyses, and tissue microarray analysis to detect novel distinct molecular aberrations in OSCC.

Introduction

Oral squamous cell carcinoma (OSCC) belongs to the 10 most common human malignancies worldwide, affecting more than 500,000 individuals per year. The 5-year survival rate for OSCC does not exceed 55% [1,2], which is mainly caused by locally aggressive tumor phenotypes [3]. Furthermore, clinicopathological parameters such as the TNM system, which are generally used as basis for therapeutic decisions, frequently fail to predict the biologic behavior of the tumors or the patients' outcome. To improve clinical management of individual patients, there is a strong requirement for a better understanding of molecular events involved in OSCC pathophysiology. Attempts to find biomarkers that identify cancerous lesions had identified several candidate genes associated with OSCC tumor progression, including p53, p16, MYC, CCND1, EGFR, and CCNL1 [4–7]. Up to now, however, no single gene was shown to have sufficient diagnostic use to predict the biologic behavior of the tumors. Although those studies contributed greatly to our current understanding, they did not explain the complexity of this malignancy.

High-throughput gene expression profiling techniques offer a unique mechanism for interrogating transcriptome-wide levels of thousands of genes expression and have proven value in defining gene expression signatures dividable in important subsets of patients, who would otherwise be undetected by conventional prognostication schemes [8,9]. Over the last few years, gene expression profiling using microarray hybridization has provided new insights in carcinogenesis and tumor cell dissemination. More than just focusing on the expression of a few genes, genomic-scale expression profiles allow the investigator to look at genetic expression variability in the context of broader genetic themes and pathways. Likewise, the expression profiles of cancers may provide the identification of specific biochemical pathways that might be targeted by new therapeutic agents.

In this study, we applied gene expression microarray technology to a collection of 35 OSCC specimens and compared the gene expression with a pool of normal oral mucosal biopsies from healthy patients. To identify pathways, which are recurrently affected by differential mRNA expression, the software package MAPPFinder was used [10–12], which allows the integration and procession of microarray data sets into the biologic context of molecular functions. To validate individual candidate genes, which were found upregulated in gene expression profiling analyses, a tissue microarray (TMA) analysis was subsequently performed in a representative collection of 306 clinically well-defined primary OSCC specimens.

Materials and Methods

RNA Expression Profiling

OSCC specimen and control samples

Thirty-five frozen tissue tumor samples were collected from patients with histologic confirmed OSCC after approval by the institutional review board of the Universitätsklinikum Heidelberg and after obtaining informed consent. Tumor samples were cut and stained by hematoxylin-eosin. Only samples containing at least 80% tumor cells were used for further analyses. Six oral mucosa samples collected from healthy donors served as control tissue. Nucleic acid extraction of high-molecular weight RNA from frozen tumor tissue was carried out by ultracentrifugation as described elsewhere in detail [13].

Transcriptome amplification and labeling

Two micrograms of total RNA from each patient and Human Universal Reference RNA (Stratagene, La Jolla, CA) were used in a T7-polymerase-based transcriptome amplification method, which was described elsewhere in detail [14].

70mere oligonucleotide microarrays

A set of 35,035 gene-specific 70mere oligonucleotide probes (Human OligoSet 4.0; Operon, Cologne, Germany) was printed on glass slides coated with epoxy-silane (Schott Nexterion, Jena, Germany). The microarray chip represents approximately 25,100 unique genes and 39,600 transcripts excluding control oligos. A variety of data sources was used to cover the genes from human mitochondrial genome, RNA genes, micro-RNA genes, the endogenous human viral genes, and the exogenous reporter genes (Operon). Microarray production, prehybridization treatment, hybridization, and posthybridization washes, and data acquisition were performed as described previously [15]. After hybridization and stringent washing, fluorescence intensity images were acquired using a dual-laser scanner (G2505 B; Agilent Technologies, Santa Clara, CA) and were analyzed with the GenePix Pro 6.0 imaging software (Molecular Devices, Union City, CA). All hybridization experiments were repeated with inversely labeled sample and reference.

Data preprocessing

Result files containing all relevant raw data were processed using the in-house-developed ChipYard microarray analysis software (http://www.dkfz.de/genetics/ChipYard/), the statistical programming language R, and packages of the Bioconductor project. Raw fluorescence intensity values were normalized applying variance stabilization. We assessed the quality of all hybridizations by generating scatter plots and gradient plots using the Bioconductor package “limma” (www.bioconductor.org). The raw and normalized data are deposited in the Gene Expression Omnibus database (http://www.ncbi.nlm.nih.gov/geo/; Accession No. GSE10121).

Statistical analysis

After normalization, all gene expression data were analyzed for differences between normal and tumor samples using analysis of variance with Bonferroni multiple testing correction. All genes with a log-value between -1 and 1 were excluded from further analyses. To identify pathways, which were likely to be affected by differentially expression, MAPPFinder, a component of Gen- MAPP software package version 2.1, was used [10–12]. MAPPFinder produces a statistically ranked list of Gene Ontology (GO) biologic categories associated with each cluster, from which the most significant nonsynonymous groups are listed. MAPPFinder analysis was performed on a set of 96 MAPPs, a representation of a biologic relationship between genes or gene products, calculating the percentage of genes meeting the criterion for each MAPP. A positive z score indicates that there are more genes meeting the criterion in a MAPP than would be expected by random chance. A negative z score indicates that there are fewer genes meeting the criterion than would be expected by random chance.

Quantitative real-time reverse transcriptase-polymerase chain reaction

To validate expression profiling data, mRNA levels of selected candidate genes TRIO, TNFRSF18, EGFR, MAPK3, and MAPK7 were tested by quantitative real-time reverse transcriptase- polymerase chain reaction (RT-PCR). For normalization purposes, mean cDNA expression levels of the housekeeping genes PGK1 and LamininB1 were measured. Details of the experimental setup are described elsewhere [16,17]. Primer sequences used in these experiments are listed in Table W1.

Tissue Microarray Analysis

Tumor material and patients' characteristics

Paraffin-embedded tumor specimen of primary OSCC were obtained from the archives of the Institute of Pathology of the University Hospital Heidelberg after approval by the local institutional review board. For all tumor samples, clinical and follow-up data of the patients were available from the Department of Oral and Craniomaxillofacial Surgery of the University Hospital Heidelberg. Mean age of the patients was 61 years at the time of diagnosis. Tumors were staged according to the TNM system of the International Union against Cancer (UICC). Of 306 tumor specimens available for TMA experiments, 143 tumors were T1/2 and 163 were T3/4 tumors. One hundred twenty-five tumors were graded according to UICC stage I to III and 207 were graded according to stage IV, respectively. One hundred twenty-three tumors presented no regional lymph node metastases at the time of diagnosis, whereas 183 exhibited regional lymph node metastases.

Tissue microarray generation

The OSCC-TMA was generated as previously described [18]. Briefly, hematoxylin-eosin.stained sections were cut from each donor block to define representative tumor regions. Small tissue cylinders with a diameter of 0.6 mm were taken from selected areas of each donor block using a tissue chip microarrayer (Beecher Instruments, Silver Spring, MD) and were transferred to a recipient paraffin block. The recipient paraffin block was cut in 5-µm paraffin sections using standard techniques. Five oral mucosa biopsies from healthy donors were incorporated in the recipient block as control specimen.

Immunohistochemistry

The immunohistochemistry (IHC) was performed using the DAKO Real Kit (DAKO; Hamburg, Germany) according to the manufacturer's protocol. Monoclonal mouse antibodies against the phosphorylated forms of ERK1 and ERK5 (Cell Signaling Technology, Beverly, MA) were used at a dilution of 1:1000 for IHC experiments. The primary antibody for phosphorylated ERK1 was the rabbit monoclonal phosphor-p44/42 mitogen-activated protein (MAP) kinase (Thr202/Tyr204) antibody that detects endogenous levels of human p42 and p44 MAP kinase (ERK1 and ERK2) only when they are phosphorylated at Thr202 and Tyr204, respectively. The primary antibody for phosphorylated ERK5 was rabbit phospho-Erk5 (Thr218/Tyr220) antibody that detects endogenous levels of human ERK5 MAP kinase only when they are phosphorylated at Thr218 and Tyr220, respectively. Evaluation of IHC experiments was based on the percentage of cells, which showed distinct nuclear and cytoplasmic staining. Oral mucosa control specimen showed slight nuclear and cytoplasmic staining, which was set as baseline immunoreactivity (0/None) in the following arbitrary score: none, <5% cytoplasmic and nuclear staining; weak, 5% to 30% cytoplasmic and nuclear staining; moderate, 31% to 60% cytoplasmic and nuclear staining; and strong, >60% cytoplasmic and nuclear staining. For statistical analyses, none and weak stainings were combined and counted as low expression, whereas moderate and strong stainings were grouped together and scored as high expression.

Statistical analysis

Nonparametric univariate analysis using chisquare test was performed to compare the prevalence of high ERK1 and ERK5 expression with T-stadium, UICC stage, and the presence of lymph node metastases of the primary tumors. For overall survival, Kaplan-Meier curves of tumors with high versus low ERK1 and ERK5 expression were analyzed by log-rank test. P ≤ .05 was considered significant. All statistical analyses were performed using R for windows version 2.4.1.

Results

70mere Microarray Expression Profiling

In the present study, n = 35 tumor specimens of primary OSCCs were analyzed for global gene expression using 70mere oligonucleotide microarrays containing 35,035 gene-specific 70mere oligonucleotides. After quality control, an analysis of variance and a following Bonferroni multitest were performed for the identification of differential gene expression between normal tissue and tumor samples. Gene expression was considered as significantly changed, if exceeding the multiple testing cutoff, computed in this case with -log10(P) of 5.65 according to the Bonferroni criterion. Using this approach, we identified 7390 genes significantly and differentially expressed between OSCC and normal oral mucosa (Table W2).

Pathway Analysis

To obtain a comprehensive view of global gene expression in OSCC, pathway analysis was applied to this data set of 7390 genes to identify molecular pathways that contained a number of altered transcript levels in OSCC compared to healthy mucosa. MAPPFinder, a component of the software package GenMAPP version 2.1 containing 96 MAPPs, was used. We ran the MAPPFinder analysis on this data set using two criteria, either an increase (fold change > 1 and -log10(P) > 5.65) or a decrease (fold change < -1 and -log10(P) > 5.65) in gene expression to obtain pathways with mostly upregulated genes and pathways with mostly downregulated genes. A pathway was defined as significantly affected by differentially expressed genes, if P ≤ .05.

In this respect, several significantly altered pathways were revealed with this approach. We found an amount of twenty-three deregulated pathways with MAPPFinder (Table 1). Of those pathways, which were upregulated in the tumor samples analyzed, several were related to inflammatory response system [B cell receptor, P < .001; interleukin (IL) 5, P < .001; IL-7, P < .001; IL-2, P < .001; tumor necrosis factor alpha (TNFα)-nuclear factor kappa B (NF-κB), P < .001; IL-6, P = .004; IL-4, P = .019; T cell receptor, P = .035]. Other pathways with a high number of differentially expressed genes were cancer-related like MAP kinase signaling (P < .001; Figure 1), transforming growth factor-beta (TGF-β) signaling (P = .015), and signaling pathways involved in apoptosis (P = .037). Pathways with significantly downregulated genes were the ribosomal protein pathway (P < .001) and the pathway of members of the electron transport chain (P < .001). The MAP kinase signaling pathway exhibited the highest number of altered genes. The altered expressed genes of this pathway are listed in Table 2.

Table 1.

Identification of Molecular Pathways with Significantly Upregulated (Above) and Downregulated Genes (Below) in OSCC Compared to Healthy Mucosa By MAPPFinder Analysis.

MAPP Name	n (Changed)	n (Measured)	n (on MAPP)	% (Changed)	% (Present)	z	P
MAPK signaling pathway_	72	151	162	47.7	93.2	4.675	0
TNFα-NF-κB	80	181	187	44.2	96.8	4.096	0
IL-2 NetPath 14	38	75	76	50.7	98.7	3.817	0
IL-7 NetPath 19	25	44	44	56.8	100.0	3.799	0
IL-5 NetPath 17	34	66	69	51.5	95.7	3.726	0
B cell receptor NetPath 12	68	155	158	43.9	98.1	3.681	0
GPCRDB_Other	11	83	119	13.3	69.7	.3.484	0
GPCRDB class rhodopsin-like	27	212	262	12.7	80.9	.5.873	0
Insulin signaling	65	155	159	41.9	97.5	3.144	.002
IL-6 NetPath 18	42	96	100	43.8	96.0	2.839	.004
p38 MAPK signaling pathway	17	32	34	53.1	94.1	2.777	.006
IL-3 NetPath 15	43	99	101	43.4	98.0	2.816	.007
TGF-β receptor	57	145	151	39.3	96.0	2.331	.015
Proteasome degradation	23	50	61	46.0	82.0	2.38	.017
IL-4 NetPath 16	27	62	62	43.5	100.0	2.233	.019
Peptide GPCRs	10	60	73	16.7	82.2	.2.37	.026
IL-9 NetPath 20	12	23	24	52.2	95.8	2.252	.032
T cell receptor NetPath 11	50	129	135	38.8	95.6	2.053	.035
mRNA processing reactome	48	121	127	39.7	95.3	2.207	.036
Apoptosis	33	80	82	41.3	97.6	2.092	.037
Fas pathway and stress induction	17	36	38	47.2	94.7	2.175	.041
Fatty acid beta oxidation 3	5	8	8	62.5	100.0	1.959	.046
Ribosomal_Proteins	35	84	88	41.7	95.5	15.668	0
Electron transport chain	21	103	105	20.4	98.1	7.309	0

Figure 1.

Overview of differentially expressed genes in 35 OSCC compared to six healthy mucosa specimen involved in MAP kinase signaling (modified from the KEGG MAPK pathway). Dark gray color of gene symbols represents the significance of difference [-log10 (P)]; the corresponding numbers are the average fold change (log fold).

Table 2.

Increased Expression of Genes Involved in MAP-Kinase Signaling Pathway in 35 OSCC Samples Compared to Six Healthy Mucosa Specimens.

Ensembl ID	Gene Symbol	Gene Name	-log10(P)	log (Fold Change)
ENSG00000100311	PDGFB	Platelet-derived growth factor B chain precursor	15.467	3.05696878
ENSG00000180370	PAK2	Serine/threonine-protein kinase PAK 2	14.294	3.807238826
ENSG00000107566	CHUK	SPFH domain-containing protein 1 precursor	14.274	4.832758228
ENSG00000114738	MAPKAPK3	MAP kinase-activated protein kinase 3	14.104	4.067114259
ENSG00000089022	MAPKAPK5	MAP kinase-activated protein kinase 5	12.030	2.129482743
ENSG00000137764	MAP2K5	Dual-specificity mitogen-activated protein kinase kinase 5	11.859	4.950684923
ENSG00000105221	AKT2	RAC-beta serine/threonine-protein kinase	11.838	5.375739428
ENSG00000154229	PRKCA	Protein kinase C alpha type	11.549	2.137398935
ENSG00000119699	TGFB3	Transforming growth factor beta-3 precursor	11.317	5.585986774
ENSG00000166484	MAPK7	Mitogen-activated protein kinase 7	11.211	2.721242004
ENSG00000105550	FGF21	Fibroblast growth factor 21 precursor (FGF-21)	11.099	1.828757871
ENSG00000167193	CRK	Protooncogene C-crk	10.696	4.945734007
ENSG00000142208	AKT1	RAC-alpha serine/threonine-protein	10.577	4.734240647
ENSG00000108861	DUSP3	Dual-specificity protein phosphatase 3	10.476	3.187181556
ENSG00000115953	PPP3R1	Calcineurin subunit B isoform 1	10.403	1.687353652
ENSG00000156711	MAPK13	Mitogen-activated protein kinase 13	9.925	3.877349348
ENSG00000085276	EVI1	Ecotropic virus integration 1 site protein	9.862	1.102867786
ENSG00000099725	PRKY	Serine/threonine-protein kinase PRKY	9.857	4.186894009
ENSG00000126458	RRAS	Ras-related protein R-Ras	9.828	3.738753854
ENSG00000142733	MAP3K6	Mitogen-activated protein kinase kinase kinase 6	9.684	4.164230763
ENSG00000112658	SRF	Serum response factor	9.667	2.390663287
ENSG00000169032	MAP2K1	Dual-specificity mitogen-activated protein kinase kinase 1	9.664	4.265539167
ENSG00000112062	MAPK14	Mitogen-activated protein kinase 14	9.647	1.394125597
ENSG00000138794	CASP6	Caspase-6 precursor	9.462	2.734601578
ENSG00000149269	PAK1	Serine/threonine-protein kinase PAK 1	9.420	2.504398448
ENSG00000106799	TGFBR1	TGF-beta receptor type I precursor	9.414	2.058273508
ENSG00000174775	HRAS	GTPase HRas precursor	9.398	5.345430439
ENSG00000104365	IKBKB	Inhibitor of nuclear factor kappa B kinase beta subunit	9.303	2.735869467
ENSG00000162889	MAPKAPK2	MAP kinase-activated protein kinase 2	9.128	3.689152094
ENSG00000109971	HSPA8	Heat shock cognate 71 kDa protein	9.061	4.628190782
ENSG00000077782	FGFR1	Basic fibroblast growth factor receptor 1 precursor	9.035	4.913382054
ENSG00000075429	CACNG5	Voltage-dependent calcium channel gamma-5 subunit	8.976	.1.9493298
ENSG00000127191	TRAF2	TNF receptor-associated factor 2	8.954	2.247565686
ENSG00000113013	HSPA9B	Stress-70 protein, mitochondrial precursor	8.893	5.378858628
ENSG00000177885	GRB2	Growth factor receptor-bound protein 2	8.777	3.249343399
ENSG00000132155	RAF1	RAF protooncogene serine/threonine-protein kinase	8.731	4.607548955
ENSG00000011485	PPP5C	Serine/threonine-protein phosphatase 5	8.703	2.290087262
ENSG00000071054	MAP4K4	Mitogen-activated protein kinase kinase kinase kinase 4	8.438	4.156174225
ENSG00000161326	DUSP14	Dual-specificity protein phosphatase 14	8.327	4.528707616
ENSG00000067191	CACNB1	Voltage-dependent L-type calcium channel beta-1 subunit	7.982	1.761273845
ENSG00000107643	MAPK8	Mitogen-activated protein kinase 8	7.942	1.56455685
ENSG00000106211	HSPB1	Heat shock protein beta-1 (HspB1)	7.838	1.440732073
ENSG00000126934	MAP2K2	Dual-specificity mitogen-activated protein kinase kinase 2	7.818	2.378935599
ENSG00000140285	FGF7	Fibroblast growth factor 7 precursor (FGF-7)	7.738	.1.472920826
ENSG00000120875	DUSP4	Dual-specificity protein phosphatase 4	7.623	2.488597455
ENSG00000105329	TGFB1	Transforming growth factor beta-1 precursor	7.621	3.393246251
ENSG00000134259	NGFB	Beta-nerve growth factor precursor (Beta-NGF)	7.586	1.28841258
ENSG00000075388	FGF4	Fibroblast growth factor 4 precursor (FGF-4)	7.549	4.415819419
ENSG00000128272	ATF4	Cyclic AMP-dependent transcription factor ATF-4	7.490	4.632776685
ENSG00000126583	PRKCG	Protein kinase C gamma type	7.399	2.201198322
ENSG00000166501	PRKCB1	Protein kinase C beta type	7.396	2.359630993
ENSG00000132906	CASP9	Caspase-9 precursor	7.367	4.394431415
ENSG00000100485	SOS2	Son of sevenless homolog 2	7.271	2.816563747
ENSG00000128340	RAC2	Ras-related C3 Botulinum toxin substrate 2 precursor	7.256	1.929198943
ENSG00000107968	MAP3K8	Mitogen-activated protein kinase kinase kinase 8	7.202	2.884100613
ENSG00000138834	MAPK8IP3	C-jun-amino-terminal kinase-interacting protein 3	7.163	2.707867082
ENSG00000123739	PLA2G12A	Group XIIA secretory phospholipase A2 precursor	7.105	.1.889796803
ENSG00000143851	PTPN7	Tyrosine-protein phosphatase nonreceptor type 7	6.948	1.921973144
ENSG00000102882	MAPK3	Mitogen-activated protein kinase 3	6.906	3.478604152
ENSG00000176697	BDNF	Brain-derived neurotrophic factor precursor (BDNF)	6.825	1.762730087
ENSG00000141480	ARRB2	Beta-arrestin-2 (Arrestin, beta 2)	6.756	2.512394619
ENSG00000085511	MAP3K4	Mitogen-activated protein kinase kinase kinase 4	6.655	1.171686631
ENSG00000186895	FGF3	Fibroblast growth factor 3 precursor (FGF-3)	6.651	1.401698475
ENSG00000115904	SOS1	Son of sevenless homolog 1	6.630	2.41943162
ENSG00000141510	TP53	Tumor suppressor p53-binding protein 1	6.479	2.802984167
ENSG00000073009	IKBKG	NF-κB essential modulator (NEMO)	6.358	2.203573687
ENSG00000170458	CD14	Monocyte differentiation antigen CD14 precursor	6.330	3.763791737
ENSG00000155903	RASA2	Ras GTPase-activating protein 2	6.092	1.297339857
ENSG00000136238	RAC1	Ras-related C3 Botulinum toxin substrate 1 precursor	6.078	.2.009429284
ENSG00000120129	DUSP1	Dual-specificity protein phosphatase 1	6.043	2.637357306
ENSG00000138032	PPM1B	Protein phosphatase 2C isoform beta	6.006	3.908525816
ENSG00000058335	RASGRF1	Guanine nucleotide-releasing protein	5.905	2.008095195
ENSG00000134853	PDGFRA	Alpha platelet-derived growth factor receptor precursor	5.808	1.093708029
ENSG00000113578	FGF1	Fibroblast growth factor 1 precursor (FGF-1)	5.786	2.292817534
ENSG00000135090	TAOK3	Serine/threonine-protein kinase TAO3	5.759	2.493439143
ENSG00000072062	PRKACA	cAMP-dependent protein kinase, alpha-catalytic subunit	5.661	1.435441441
ENSG00000104814	MAP4K1	Mitogen-activated protein kinase kinase kinase kinase 1	5.626	2.539747316
ENSG00000142875	PRKACB	cAMP-dependent protein kinase, beta-catalytic subunit	5.604	2.081661093

For each transcript, the mean value of these six mucosa specimens was subtracted. An increased expression was defined if this value was higher than three standard deviations of these six healthy mucosa specimens.

Quantitative Real-Time RT-PCR

To confirm the findings of the microarray analysis, we performed quantitative real-time RT-PCR using primers specific for TRIO, TNFRSF18, EGFR, MAPK3/ERK1, and MAPK7/ERK5. The fold differences in expression between tumor specimen and control mucosa specimen predicted by 70mere microarray expression profiling were compared to those fold differences obtained by RT-PCR (see Figure W1).

Quantitative PCR confirmed the direction of fold change in 62 (84.9%) of 73 analyses. Overall, these results confirm our findings of differential gene expression by microarray analysis.

Tissue Microarray Analyses

From those pathways found deregulated by cDNA microarray expression profiling in the OSCC tumor samples, further analyses were focused on the MAP kinase signaling pathway, because MAP kinase signaling is supposed to be critically involved in a variety of cancerspecific functions, such as proliferation, dedifferentiation, and evasion from apoptosis. For TMA analysis, antibodies against ERK1 and ERK5, which play a central role in signal transduction in the MAP kinase signaling pathway, were selected. To test whether OSCC also show an increase of ERK1 and ERK5 protein expression, an immunohistochemical analysis was performed on TMA sections (Figure 2). To assess the functional properties of these proteins in the primary tumor tissue, antibodies against the active phosphorylated form of ERK1 and ERK5 were used. The overall frequency of high pERK5 expression was 27.4% (79 of 277). There was a significantly higher prevalence for high pERK5 expression in T1/2 versus T3/4 tumors (P = .015), in Stage I to III versus Stage IV tumors (P = .008), and for tumors with lymph node metastasis (N1-3) versus tumors without lymph node metastasis (N0, P = .016). For high pERK1 expression, the overall frequency was 22.8% (48 of 209) without obtaining any correlation of high pERK1 expression with the clinical parameters analyzed. All data obtained from TMA analyses are shown in Table 3. Kaplan-Meier analysis revealed no difference in the overall survival for tumors with high expression versus low/no expression of the proteins analyzed (P > .05, data not shown).

Figure 2.

Detection of differential pERK1 (left side) and pERK5 (right side) protein expression on TMA sections as detected by IHC. Normal oral mucosa specimens (A and D; original magnification, x20) exhibit a slight pERK1 and pERK5 staining. For OSCC specimen, two examples of absent (B and E; original magnification, x20) and high (C and F; original magnification, x20) ERK1 and ERK5 protein expression are shown.

Table 3.

Frequency of Overexpressed ERK1 and ERK5 in Clinically Defined Tumor Subgroups.

HNSCC (N = 306)	n	ERK1	ERK5
T1/2	143	18.3% (20/109)	20.3% (27/133)
vs
T3/4	163	28% (28/100)	34% (49/144)
P		.136	.015
N0	124	18.2% (18/99)	19.3% (22/114)
vs
N1-3	182	27.3% (30/110)	33.1% (54/163)
P		.163	.016
Stages I–III	123	18.0% (16/89)	17.9% (19/106)
vs
Stage IV	183	26.7% (32/120)	33.3% (57/171)
P		.19	.008

P values for univariate subgroup analysis are added. P values ≤ .05 are considered as significant.

Discussion

Gene expression profiling to OSCC specimen using different types of cDNA arrays has been performed in several studies to define distinct expression signatures for specific clinical stages [19,20]. To actually understand the process of OSCC progression and metastases, however, it would be essential to know which biologic aberrations are represented by these classifying genes and to what extent they contribute to tumorigenesis. To obtain a comprehensive overview of activated signal transduction pathways in biologic systems, several programs have been developed, which allow the integration and procession of microarray data sets into the context of molecular functions and interrelated dependencies of annotated genes [21–23]. In contrast to those approaches, however, which mainly focus on welldefined metabolic pathways, the program MAPPFinder, which was used in the present analysis, is closely linked to a broader base of pathway information provided by the GO Consortium [11,12]. MAPPFinder dynamically links gene expression data to the GO hierarchy at the level of biologic processes, thereby making it a valuable tool to define novel aberrant pathways in homogenous cohorts of tumor samples [24].

In the present study, a whole-transcriptome expression analysis was performed for 35 OSCC specimens resulting in 7390 differentially expressed transcripts compared to a pool of healthy mucosa samples. By subsequent pathway analysis using MAPPFinder, 24 aberrantly expressed molecular pathways were detected. Among the cancerrelated pathways in this data set, most distinct aberrations—mostly upregulations—were found in the MAP kinase signaling network. The MAP kinase signaling pathways are involved in various cellular functions, including cell proliferation, differentiation, and migration by the activation of protooncogenes such as JUN, FOS, MYC, and ELK1 (Figure 1). FourMAP kinase signaling pathways have been identified, namely, the ERK1 pathway, the c-jun N-terminal-regulated kinase (JNK) pathway, the p38 pathway, and the ERK5 pathway [25]. Whereas the JNK and the p38 pathway are signaling cascades, which are mainly stress-activated by proinflammatory cytokines, ERK1 and ERK5 are additionally induced by epidermal growth factor receptor (EGFR) activation [26]. Because EGFR activation is known to be decisively involved in head and neck squamous cell carcinoma (HNSCC) initiation and progression [27], subsequent analyses were focused on ERK1 and ERK5 signaling pathways in the present study. High ERK1 expression and an association with tumor progression and adverse clinical outcome had been observed in several tumor entities, e.g. in primary hepatocellular carcinoma [28], cholangiocarcinoma [29], breast cancer [30], and non-small cell lung cancer [31]. For HNSCC, an involvement of ERK1 signaling in angiogenic processes through vascular endothelial growth factor activation was postulated [32]. In our present TMA analysis, high ERK1 protein expression was found in about 20% of primary OSCC tumor samples. Surprisingly, however, an association with clinical parameters such as advanced UICC stage and presence of lymph node metastases was not found for high ERK1 but for high ERK5 protein expression, although ERK1 and ERK5 coexpression was frequently found. Mitogen signal-regulated ERK5 overexpression was shown to be associated with metastatic prostate cancer [33], but for HNSCC or OSCC, a participation of ERK5 expression in tumor progression has not yet been described. Analyses of tumor cell systems, however, suggested that MAP kinase signaling through ERK5 is a distinct molecular pathway, which might regulate cellular functions such as the activation of protooncogenes originally attributed to ERK1 [34]. Furthermore, it could be shown that ERK5 but not ERK1 signaling contributes to Src-mediated disruption of actin cytoskeleton organization [35]. Because the disorganization of the cytoskeleton is known to be one of the initial steps that are required for the development of metastases, this observation is in concordance with our data of a correlation of high ERK5 expression and the presence of lymph node metastases in OSCC. Therefore, one might speculate that ERK5 signaling is more important in the EGFR-mediated metastasizing process in OSCC than in the classic ERK1 signaling pathway.

A further distinct finding in signaling pathway analysis of the expression profiling data obtained from the OSCC collection was frequently upregulated pathways involved in inflammatory response, e.g., B cell receptor signaling, TNFα signaling, IL-2 signaling, IL-5 signaling, and T cell receptor signaling (Table 1). In general, it is a well-investigated phenomenon that inflammation plays an important role in tumor promotion, particularly in HNSCC development [36]. It was recently postulated that tumor cells can modify the surrounding stroma through the production of cytokines and growth factors and that this locally changed host microenvironment influences the proliferative and invasive behavior of tumor cells [37]. Particularly for HNSCC, it has been demonstrated that primary tumors express a variety of proinflammatory cytokines, including IL-1a, IL-6, IL-8, and granulocyte macrophage colony-stimulating factor that may attract immune effector cells to the tumor microenvironment [38]. In this context, one might interpret the results of the pathway analysis, which showed an increased expression of proinflammatory molecules in the tumor biopsies, as a distinct tumor-specific feature involved in the pathogenesis of the infiltrating process. Conversely, however, inflammatory cytokines are also expressed by cells of the immune system themselves during such an inflammatory process. Therefore, the inflammatory cell infiltration at the invasion front of the tumor might be the source of high cytokine expression as well. Although only tumor samples containing at least 80% tumor cell load in the biopsy were used for expression profiling analysis in the present study, a contamination with cells of the immune system cannot be definitely excluded. Nevertheless, with the development of more efficient microdissection techniques and RNA amplification protocols, a whole-transcriptome expression profiling with smaller RNA samples would be possible. Then, a comprehensive signaling pathway analysis of the expression of immunomodulatory molecules might be a promising approach to further define the role of inflammation in OSCC progression.

Furthermore, pathway analyses showed a significant upregulation of apoptosis-related genes. Although evading apoptosis is a hallmark of almost all malignant tumors, little is known about the actual role of apoptosis-related genes in OSCC development. Previous analyses, for example, suggested increased levels of the anti-apoptotic protein bcl-2 in OSCC samples [39–41]. A recent study, however, using an animal model system exhibiting chemically induced OSCC, found a high expression of the proapoptotic protein bax but decreased levels of bcl-2 in early oral cancer specimens [39]. Another study on advanced primary OSCC showed an association of high expression of the antiapoptotic protein survivin, with favorable outcome of the patients [42]. Tissue microarray technology might be helpful to further delineate the precise role of apoptosis-related genes in OSCC initiation and progression.

Whereas most of the aberrant signaling pathways were identified as upregulated, the expression of ribosomal proteins was found as significantly downregulated in the tumor collection analyzed. Furthermore, MAPPFinder analysis defined the highest z value for ribosomal protein signaling of all pathways analyzed, indicating that its downregulation might be a decisive aberration in OSCC specimen compared to oral mucosa specimen. Only a few published studies have evaluated the role of cytoplasmic ribosomal proteins in carcinogenesis. Significant changes in the expression of several ribosomal proteins have been reported to occur in colorectal carcinoma [43]. In ovarian tumor cell lines, ribosomal proteins, such as S8, S24, and L32, are much more abundantly expressed in differentiated tumor cells than in lessdifferentiated ones [44]. For OSCC, a downregulation of some ribosomal proteins were found in a previous cDNA expression profiling study, with the downregulation of ribosomal protein S13 discriminating between metastatic and nonmetastatic OSCC in small tumor collection [45]. Because of the observation that a large number of genes encoding ribosomal protein are downregulated in OSCC in our study, further detailed investigation to prove the significance of such global downregulation is warranted.

Pathway signaling analyses results not only contribute to a better understanding of molecular and cellular functions and the definition of potential biomarkers but also open the gate to novel molecular targets for specific therapeutic approaches. For ERK1, the negative effect on cell proliferation using specific ERK1 inhibitors has been shown in several tumor systems. In metastasizing neuroblastoma, ERK1 inhibition by the bisphosphonate zoledronic results in a decrease in tumor cell proliferation and an increase in tumor cell apoptosis [46]. For papillary thyroid carcinoma carrying an ERK1-activating BRAF mutation, specific ERK1 inhibition resulted in tumor cell growth arrest [47]. Similar effects of tumor cell growth inhibition after blocking ERK1-mediated signaling were found in cholangiocarcinoma [48] and hepatocellular carcinoma [49]. For HNSCC, targeted therapy has been recently established in clinical management by the use of the monoclonal antibody cetuximab that selectively inhibits EGFR. Although cetuximab has proven antitumor activity as a single agent and in combination with radio- and chemotherapy [50], a significant number of patients do not adequately respond to cetuximab treatment. Therefore, it has been postulated that a combination therapy with the additional specific inhibition of downstream mediators of EGFR signaling might be a useful approach to increase cetuximab response [27,51]. In this context, according to the data from the present study, ERK5, as a downstream mediator of EGFR, which is associated with advanced tumor stage, might be a novel molecular target. The use of specific ERK5 inhibitors to block EGFR-induced tumor cell proliferation might be a promising approach to support antitumor activity of cetuximab in HNSCC treatment.

In conclusion, the combined experimental approach of wholetranscriptome expression profiling, automated signaling pathway analysis and protein expression analyses by tissue microarrays resulted in the rapid definition of molecular components, which are critically involved in tumor progression. It allowed the evaluation of diagnostic and prognostic biomarkers as well as the definition of novel promising therapeutic targets. ERK5-mediated signaling of EGFR activation might be more important for OSCC progression than ERK1-mediated signaling, suggesting ERK5 as a potential interference point in future therapeutic approaches.